Steve Sakoman <sakoman@xxxxxxxxx> writes: > On Wed, Dec 23, 2009 at 1:06 PM, Robert Nelson <robertcnelson@xxxxxxxxx> wrote: >> On Wed, Dec 23, 2009 at 2:59 PM, Steve Sakoman <sakoman@xxxxxxxxx> wrote: >>> I'm seeing some strange behavior with my console serial port with 2.6.32. >>> >>> Output seems fine -- no errors in the boot messages. >>> >>> However it seems that quite often the first character typed in a >>> console session is garbled -- subsequent characters are OK: >>> >>> root@omap3:~# �s -l >>> -sh: �s: not found >>> >>> If commands are typed in rapid succession this issue doesn't show up, >>> but if several seconds pass since the last keystroke the first >>> character is garbled once again. >>> >>> Anyone else seeing this? A search of this list didn't turn anything >>> up, but perhaps my google mojo is just bad today. >>> >>> Steve >> >> Hey Steve, >> >> I saw something similar while testing 2.6.32, by chance are you using >> any power management patches? >> >> I had to disable CONFIG_CPU_IDLE: >> >> -CONFIG_CPU_IDLE=y >> -CONFIG_CPU_IDLE_GOV_LADDER=y >> -CONFIG_CPU_IDLE_GOV_MENU=y >> +# CONFIG_CPU_IDLE is not set > > Now power management patches, but CPU_IDLE is indeed enabled. I'll > give it a try with CPU_IDLE disabled. As of 2.6.32, much of the PM patches are in mainline now, including CPUidle. > Does this just work around a bug that ought to be fixed, or is > enabling CPU_IDLE just a bad idea? CPUidle itself is just fine. You've discovered a PM "feature." ;) I'm guessing you (or your userland) has enabled sleep-while-idle. Check the value of <debugfs>/pm_debug/sleep_while_idle. This enables PM features during idle. What you're seeing is a side-effect of allowing the UARTs to be idled and un-clocked during idle. The UARTs have a SW inactivity timer (default 5 secs) after which their clocks will be cut during idle. The UARTs themselves are also configured as wakeup sources by default, so any UART input will trigger a wakeup. However, the first char serves as the wakeup but the OMAP does not wakeup fast enough to actually latch the first character, resulting in either garbage or a lost char. The simplest way to avoid the problem (but implies increased power consumption) is to simply disable the UART inactivity timer for your console UART. This way, that UART will never timeout and its clocks will never be cut: # echo 0 > /sys/devices/platform/serial8250.N/sleep_timeout (where N = [0, 1, 2], as in ttySN) Another alternative (although not a very practical one) is to use something else as a wakeup source (GPIO, twl4030 keypad) before typing on the console. Personally, after so much PM development, I've now develop the habit of typing the first character of commands twice. :) Basically, there is currently no way of avoiding this problem if you are using UARTs as the wakeup source and want power savings. This will also have obvious and serious PM implications for non serial-console devices connected to UARTs (IrDA, bluetooth, modems, etc.) I've heard some ideas of muxing the UART into GPIO mode when idle and having a bitbanging to get the first char so it is not lost, but haven't seen any patches to do it. While it doesn't really solve your problem, I hope it at least helps understand the reason for it. Kevin -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
